Abstract CFD simulations of gas–liquid–solid fluidized beds have been performed in a full three-dimensional, unsteady multiple-Euler framework by means of the commercial software FLUENT. The simulation results were compared with experimental data obtained from laboratory scale three-phase fluidized beds. The significance of implementing accurate numerical schemes, as well as the choice of available k–ε turbulence models (standard, RNG, realizable), solid wall boundary conditions and granular temperature models were investigated. The results indicate that in order to minimize numerical diffusion artifacts and to enable valid discussions on the choice of physical models, third-order numerical schemes need to be implemented. The realizable turbulence formulation was unable to produce the expected solids gulf-stream pattern (i.e., rising solid particles in the core and descending solid particles near the wall) in three-phase fluidized beds whereas the RNG and standard k–ε models were able to better capture depictions of flow patterns. The best prediction of flow characteristics was obtained with a laminar model formulation accounting for the solid phase viscosity and the molecular viscosities of the two fluids.